Predictive imaging marker of bevacizumab efficacy: perfusion MRI

Intuitively it seems a “killer app” for perfusion imaging (Table 1) would be its use as an early marker for glioblastoma (GB) therapy response and relapse in the setting of anti-angiogenic therapy. In fact, in a letter commenting on the first paper characterizing bevacizumab treatment of brain tumors, the potential for perfusion imaging in monitoring anti-angiogenic therapy was raised.^1,2 The hope was that tumors that showed decreased perfusion after initiation of bevacizumab would do better than those without change. Another possibility was that perfusion would increase in tumors as they start to develop resistance to anti-angiogenic therapy. This would precede significant tumor growth, thereby adding value to standard assessment of disease burden based on tumor size measurements. While inherently appealing, neither of these hypotheses is well supported to date, despite the widespread use of bevacizumab and availability of perfusion imaging for over a decade. Potentially the issue is impacted by lack of substantially improved outcomes for patients treated with bevacizumab. After all, it is hard to have a marker of survival benefit, for instance, when little or no benefit exists. But even the pathophysiology of these tumors suggests reasons why changes in perfusion may not be tightly linked to outcomes. For instance it is not uncommon for a tumor that appears to respond to bevacizumab to show progression on the subsequent scan. Rapid development of bevacizumab resistance is not uncommon and may be independent of initial response.³ If true, this would degrade the performance of perfusion changes as an early response marker.

Table 1.

Common MRI perfusion techniques

Name	Abbreviation	Most Commonly Derived Metric	Requires Intravenous Contrast?
Dynamic susceptibility contrast	DSC	Relative cerebral blood volume (rCBV)	Yes
Dynamic contrast enhanced	DCE	Ktrans (vascular permeability surrogate)	Yes
Arterial spin labeling	ASL	Cerebral blood flow (CBF)	No

Data indicating that changes in perfusion may be useful in stratifying outcome in patients treated with anti-angiogenic therapy are not entirely absent from the literature, however. Seminal work by Sorensen et al⁴ did find a potential link between vascular permeability and outcomes in high-grade glioma patients treated with the anti-angiogenic drug cediranib: patients with tumors that showed decreased vascular permeability one day after treatment initiation did better than those that showed no such “vascular normalization.” The fact that this imaging marker was measurable only one day after the start of therapy generated much excitement for its potential to quickly identify responders. Indeed it was suggested that early perfusion changes (within days of therapy initiation) might be valuable in determining response, whereas images acquired at the usual follow-up time point (4–6 wk post therapy) may no longer be informative. Unfortunately it was subsequently shown that perfusion metrics from even early post-therapy time points do not stratify outcomes in patients treated with bevacizumab. Indeed, 2 studies evaluated changes in blood volume and changes in vascular permeability in patients with recurrent high-grade glioma treated with bevacizumab, but found no association with either progression-free or overall survival (PFS or OS, respectively), even when initial evaluation of treatment effect was performed within 3–4 days of the start of therapy.^5,6

Although overall the effects on prolonging survival in bevacizumab-treated patients is modest at best,^7,8 it is still unclear whether there is not a more substantial positive effect in a subset of patients, potentially identifiable by imaging markers. Thus investigators have started to examine the potential of perfusion imaging as a predictive rather than early response marker, with an eye toward selecting patients for bevacizumab therapy who will be most likely to benefit. Data from at least 2 preliminary studies have supported the use of perfusion metrics for this purpose, as they demonstrated an association between baseline (pretreatment) relative cerebral blood volume (rCBV) and OS in patients with high-grade gliomas treated with bevacizumab.^5,9

In the current issue of Neuro-Oncology, Kickingereder et al¹⁰ report outcomes from a much more robust analysis of this potential relationship by examining recurrent GB in 2 large cohorts, only 1 of which received bevacizumab. The authors determined that baseline rCBV values could stratify both PFS and OS, but only in the bevacizumab-treated cohort (n = 71), confirming the predictive rather than prognostic nature of this marker. High baseline rCBV was associated with short PFS and OS. In fact, patients with high baseline rCBV had less than half the median survival of patients with low baseline rCBV, with survival times similar to those of patients who were not treated with bevacizumab. Relative CBV remained predictive of survival even at first follow-up (whereas the change in rCBV was not). In both bevacizumab-treated and untreated cohorts, contrast-enhancing and fluid attenuated inversion recovery volumes at first follow-up were also significantly associated with PFS and OS, supporting standard assessment paradigms (eg, Response Assessment in Neuro-Oncology criteria) as informative measures that correlate with outcomes.

The authors also employed a potentially important technical advance: the use of an arterial input function that allowed for determination of rCBV without normalization to white matter or a contralateral region of interest, steps that can generate error and degrade reproducibility. To date, the application of perfusion imaging has been hampered by lack of standardization, particularly in postprocessing methods, and the best method of normalization is debatable. Using an automatically generated arterial input function as the authors did may increase reproducibility that will be critical for future multicenter trials required to rigorously validate baseline rCBV and other perfusion biomarkers.

Another point of interest of the Kickingereder et al paper is that the predictive accuracy for baseline rCBV was fairly impressive: 82% for 6-month PFS and 79% for OS. A high degree of accuracy is crucial for the implementation of imaging biomarkers in GB therapy, due to the limited treatment options for patients with progressive recurrent disease. For instance, even if a patient has a small chance of responding to it, bevacizumab may still be the best treatment option, because no other proven effective therapy exists. Thus what predictive accuracy is necessary to achieve clinical impact for drug treatment selection needs to be carefully considered. One way to achieve this goal might be to combine predictive markers. The most studied and well characterized predictive marker for bevacizumab treatment efficacy is derived from diffusion imaging.^11,12 As Kickingereder and colleagues suggest, the combination of these perfusion and diffusion metrics may improve accuracy to the point where clinical utility is achieved. Such an ability to successfully select patients for bevacizumab (or other specific therapy) would be a milestone in the application of imaging to GB management.